obra/superpowers

What it solves

Coding agents often jump straight into writing code without a clear plan or testing strategy, leading to errors and deviations from the user's actual intent. Superpowers provides a structured software development methodology that forces agents to brainstorm, plan, and test before implementing features.

How it works

Superpowers is a set of composable skills and instructions that can be installed as plugins for various coding agents (such as Claude Code, Cursor, and Gemini CLI). It replaces ad-hoc coding with a mandatory workflow:

1. Brainstorming: The agent asks questions to extract a specification and creates a design document.
2. Workspace Setup: It uses Git worktrees to create isolated development branches.
3. Planning: It breaks the work into small, verifiable tasks.
4. Implementation: It uses either subagent-driven development (dispatching fresh agents per task) or inline execution.
5. TDD: It enforces a Red-Green-Refactor cycle where tests are written before code.
6. Review: It performs code reviews against the plan before finishing the branch.

Who it’s for

Developers using AI coding agents who want more reliable, professional engineering practices (like TDD and systematic debugging) integrated into their AI-driven development process.

Highlights

• Cross-Agent Compatibility: Supports a wide array of harnesses including Claude Code, Cursor, Devin, GitHub Copilot CLI, and others.
• Subagent-Driven Development: Ability to dispatch parallel agents to handle specific tasks with a two-stage review process.
• TDD Enforcement: Strictly mandates writing failing tests before writing minimal code to pass them.
• Systematic Debugging: Includes a 4-phase root cause process to move beyond guessing.
• Self-Diagnostic Tool: A dedicated skill for agents to analyze their own session transcripts to figure out why a workflow failed.